EUROSPEECH 2001 Scandinavia
Most speaker verification systems wait to collect a complete utterance from a speaker before making a decision. Faster verification can be achieved if decisions are made sequentially on smaller "chunks" of data. In this paper we present a sequential decision making algorithm in a connected digit application and discuss its properties. We show that sequential decisions, apart from requiring shorter utterances and fewer computations on the average, add another dimension of flexibility over current systems: within some limitations, they provide the ability to systematically tradeoff between the performance of the system and the amount of data needed to make a decision. Thus they make a speaker verification system work faster and be more flexible in real applications.
Bibliographic reference. Surendran, Arun C. (2001): "Sequential decisions for faster and more flexible verification", In EUROSPEECH-2001, 763-766.